Revisiting the Systematicity in Negation in the Era of In-Context Learning

2026-06-15

AI Summary

Large language models still exhibit systematic difficulties in understanding negation, particularly in identifying the scope of negation. This study addresses this challenge by integrating behavioral and representational perspectives, employing in-context learning, functional vector analysis, and representation probing to evaluate the models’ ability to recognize negation expressions and their scopes under varying output formats. The findings reveal that while models can partially detect negation cues, their performance in scope identification remains limited and is significantly influenced by output format. Although functional vectors effectively capture negation-related signals, they struggle to consistently construct representations sufficient for accurate scope determination. This work highlights the critical role of output format in negation comprehension and offers new insights into the representational mechanisms underlying negation semantics in large language models.

Abstract

Understanding the meaning of negated sentences remains one of the challenges for language models, even in the era of large language models (LLMs). We analyze systematicity regarding LLM understanding of negation from two perspectives: behavioral systematicity and representational systematicity. For behavioral systematicity, we confirm that through demonstrations and in-context learning, LLMs can recognize negation expressions and scope within sentences to some extent, but they fail to achieve perfect performance. In particular, the difficulty of the negation scope recognition for models varies depending on the output format. For representational systematicity, we analyze the extent to which function vectors can be robustly constructed from in-context examples for tasks that are essential to understanding negation. The experiments suggest that while function vectors can be composed for negation cue extraction tasks, extracting function vectors for recognizing scope is more challenging.

Problem

• negation
• systematicity
• large language models
• in-context learning
• scope recognition

Innovation

• negation scope
• in-context learning
• function vectors
• systematicity
• large language models